Abstract
This study examined how amyloid burden affects structural and functional brain network topology in subjective cognitive decline (SCD), a risk condition for Alzheimer's disease (AD). Functional and structural brain networks were analyzed in 100 individuals with SCD and 86 normal controls (NC; both sexes included) using resting-state functional MRI and diffusion tensor imaging. Topological properties of brain networks were evaluated as indicators of information exchange efficiency and network robustness. Amyloid burden in 55 SCD participants was measured using amyloid PET imaging and a frequency-based staging method, which defined global and regional amyloid burden for four anatomical stages. Compared with NC, individuals with SCD exhibited increased functional nodal efficiency and structural nodal betweenness in the left anterior and median cingulate gyri, with no differences in network-level properties. Amyloid staging revealed four cortical divisions: Stage 1, fusiform and lateral temporal gyri; Stage 2, occipital areas; Stage 3, default mode network (DMN), midline brain, and lateral frontotemporal areas; and Stage 4, the remaining cortex. The global and regional amyloid burdens of each cortical stage were positively associated with the node-level properties of a set of DMN hubs, with the left anterior and posterior cingulate gyri being congruently associated with all amyloid stages. These findings suggest that amyloid burden continuously influences network adaptations through DMN hubs, irrespective of local proximity to pathology. Increased nodal properties in cortical hubs may reflect heightened information-processing demands during early amyloid deposition in this population at risk for AD.
Significance Statement
Amyloid spreads throughout the cortex in Alzheimer’s disease (AD). It is unclear whether early amyloid deposition may trigger system-level network reorganization in individuals with subjective cognitive decline (SCD) who are at risk for AD. We examined the brain topology alterations in SCD and its relationship with amyloid deposition at different cortical stages. We found increased node-level topological properties in the core default mode network region (i.e., the cingulate cortex) in SCD. Increasing regional amyloid load at all stages showed consistent associations with the increasing node-level topological properties of the cingulate cortex in SCD. Our findings suggest that amyloid deposition impacts the system-level network adaptation via the cingulate cortex already at the very early stage and is unlikely to have a local effect in this AD-risk population.
Introduction
Neural network changes have been frequently observed during the neurodegenerative process along the Alzheimer's disease (AD) continuum (Jones et al., 2016; Yu et al., 2021). Amyloid pathology begins to accumulate in the asymptomatic preclinical stage of the disease, spreads slowly throughout the cortex over decades, and approaches a plateau in the clinical stages with manifest cognitive deficits (Jack et al., 2013). Amyloid deposition may cause local neuronal damage or disrupt fiber tracks (Teipel and Grothe, 2016), leading to altered brain networks (Yu et al., 2021) that spatially overlap with molecular-level pathological changes (Buckner et al., 2005). Alternatively, molecular-level amyloid deposition may trigger system-level network reorganization that is independent of local physical proximity to the pathology (Jones et al., 2016). Graph-theoretic analysis, used to study system-level topological properties of brain networks, has suggested a loss of highly connected areas (hubs) in AD patients at the symptomatic stages (Tijms et al., 2013). However, network changes vary across the AD continuum ranging from the asymptomatic to symptomatic stages (Hasani et al., 2021), with inconsistent results attributed to the different disease stages studied, the variable study quality, and the different methodologies (Hasani et al., 2021).
Subjective cognitive decline (SCD), defined by persistent self-perception of cognitive decline in old adults with normal cognition (Jessen et al., 2014), is a risk condition for AD (Jack et al., 2018; Jessen et al., 2020). Although a significant portion of individuals with SCD will not progress to dementia, SCD with AD pathology has been associated with pronounced longitudinal cognitive decline as compared with controls (Jessen et al., 2023; Shao et al., 2024) and has been recognized as a cognitive transition stage of AD continuum (Jack et al., 2018). Previous studies of brain network changes in SCD have yielded inconsistent results. Previous PET studies in SCD have focused on the relationship between global amyloid burden and network characteristics (Chiesa et al., 2018, 2019; Franzmeier et al., 2021; Jiang et al., 2023). However, the global amyloid burden-related changes do not necessarily reflect the effect of local amyloid burden. For instance, right frontoparietal control network connectivity was negatively associated with both global amyloid burden and local amyloid retention in the frontal lobe but positively related to local retention in the parietal lobe, indicating that different spatial extensions of amyloid deposition may affect the network alterations differently (Elman et al., 2014). Given that amyloid deposition spreads throughout the cortex following a specific spatial pattern (Fantoni et al., 2020), the current work adopts the view that system-level network changes may not be driven by the amyloid deposition in a discrete cortical region but may be collectively influenced by a set of spatially adjacent areas at a particular amyloid stage.
Using frequency-based staging methods, both autopsy (Braak and Braak, 1991) and amyloid PET studies (Grothe et al., 2017; Jelistratova et al., 2020; Levin et al., 2021) demonstrated that the deposition begins in the basal part of the brain and progresses to widespread isocortex areas, medial temporal lobe, and subcortical areas. Notably, a substantial proportion of SCD was classified below the global amyloid cutoffs but showed initial amyloid deposition in the basal part of the brain (Sakr et al., 2019; Teipel et al., 2020). As subthreshold amyloid pathology has previously been observed to affect the brain network (Palmqvist et al., 2017), it is likely the initial and subthreshold amyloid deposition may also be related to network adaptation in SCD. Previous studies have not evaluated network adaptations in relation to amyloid stages.
This study aims to explore the topological property changes of functional and structural networks in SCD and their association with amyloid stages. Topological properties were derived from resting-state functional MRI (rs-fMRI) and diffusion tensor imaging (DTI) data and compared between the SCD and the normal controls (NC). In a subset with amyloid PET, amyloid progression was characterized using a frequency-based staging method, and we further explored patterns of topology–amyloid associations across cortical stages.
Methods and materials
Participants
The present study analyzed data of 100 SCD and 86 NC obtained from the ongoing Sino Longitudinal Study on Cognitive Decline (SILCODE; ClinicalTrials.gov identifier, NCT03370744 from December 2015 to May 2021; Li et al., 2019). SCD is recruited in accordance with the international consensus of research criteria (Jessen et al., 2014): (1) age ≥60 years, right-handed, and Mandarin-speaking subjects; (2) presence of self-perceived continuous cognitive decline compared with previous normal status and unrelated to an acute event; (3) concerns (worries) associated with memory complaint; (4) cognitive tests for language, episodic memory, and executive function within normal range; and (5) failure to meet the criteria for MCI according to the Jak/Bondi criteria (Bondi et al., 2014) or dementia. The NC group was within the normal range on cognitive tests and reported no self-perceived cognitive declines of concern. The exclusion criteria are as follows: (1) history of stroke; (2) current major psychiatric diagnoses such as severe depression and anxiety (individuals with subclinical depressive or anxious symptoms are not excluded); (3) other neurological conditions that could cause cognitive decline (e.g., brain tumors, Parkinson's disease, encephalitis, or epilepsy) rather than AD spectrum disorders; (4) other diseases that could cause cognitive decline (e.g., thyroid dysfunction, severe anemia, syphilis, or HIV); (5) history of psychosis or congenital mental developmental delay; (6) cognitive decline caused by traumatic brain injury; and (7) inability to complete the study protocol or the presence of contraindications for MRI.
All participants performed within the normal range on standardized neuropsychological tests and received MRI (T1-weighted structural, rs-fMRI, and DTI images) and apolipoprotein E (ApoE) genotype assessments. A subset of 55 SCD underwent amyloid PET imaging. The study was approved by relevant ethics committees and radiation protection authorities. All participants provided written informed consent (Jiang et al., 2023).
Clinical and cognitive assessments
All subjects underwent standardized neuropsychological assessment probing episodic memory, executive function, and language, as previously reported (Jiang et al., 2023). A structured Subjective Cognitive Decline Interview (Miebach et al., 2019) was conducted to obtain the number of cognitive domains (i.e., memory, language, executive function, attention, and others) affected (SCD-domain score), as well as the presence of features associated with increased risk of cognitive decline (SCD-plus score). The SCD-plus features include (1) perceived decline in memory, (2) onset of SCD within the last 5 years for individuals over 60, (3) concerns related to SCD, and (4) perception of cognitive decline compared with peers (Miebach et al., 2019; Jiang et al., 2023). General cognitive performance was obtained from the mini-mental state examination, and episodic memory was obtained from the auditory verbal learning test (Zhao et al., 2012). In addition, for measuring psychiatric symptoms, the Hamilton depression scale (HAMD; Williams, 1988), the 15-item short form of the geriatric depression scale (GDS; Xie et al., 2015), and the Hamilton anxiety scale (HAMA; Maier et al., 1988) were implemented.
MRI and PET imaging
Imaging acquisition
High-resolution T1–weighted MRI acquired on a 3 T GE Signa PET/MR scanner (GE Healthcare) with a spoiled gradient-recalled sequence with the following parameters: field of view (FOV), 256 × 256 mm2; slice thickness, 1 mm (no gap); a total of 192 slices; repetition time (TR), 6.9 ms; echo time (TE), 2.98 ms; inversion time (TI), 450 ms; flip angle (FAn), 12°; and voxel size, 1 × 1 × 1 mm3.
Rs-fMRI was acquired using an echo-planar imaging (EPI) sequence on the same 3 T GE Signa PET/MR scanner: FOV, 224 × 224 mm2; slice thickness, 4 mm (gap, 1 mm); a total of 28 slices and 240 volumes; TR, 2,000 ms; TE, 30 ms; FAn, 90°; and voxel size, 3.5 × 3.5 × 5 mm3.
DTI was acquired using a single-shot spin-echo diffusion–weighted EPI sequence: FOV, 224 × 224 mm2; matrix, 112 × 112; slice thickness, 2 mm (no gap); a total of 70 slices (interleaved order); TR, 16,500 ms; TE, 95.6 ms; 30 gradient directions; 5 b0 images (b = 1,000 s/mm2); and voxel size, 2 × 2 × 2 mm3.
For PET, two different scanning protocols were applied using a GE Signa and GE Discovery PET/CT Elite scanner. For the GE Signa PET scanner, the tracer activity was between 259 and 370 MBq (5.55–7.4 MBq/kg), and the emission scan started after 40 min postinjection with following acquisition parameters: single 35 min time frame; reconstruction algorithm, VPFXS; eight iterations; 32 subsets; matrix, 192 × 192; in-plane pixel resolution, 1.82 × 1.82 mm; 89 slices; and slice thickness, 2.78 mm. For the GE Discovery PET/CT Elite scanner, the tracer activity ranged from 259 to 370 MBq (5.55–7.4 MBq/kg), with the emission scan starting between 50 and 70 min postinjection: single 10 min time frame; reconstruction algorithm, VPFXS; five iterations; 18 subsets; matrix, 192 × 192; in-plane pixel resolution, 1.56 × 1.56 mm; 47 slices; and slice thickness, 3.27 mm.
MRI data preprocessing
We preprocessed T1 and rs-fMRI data using SPM12 (fil.ion.ucl.ac.uk/spm/). T1 images were segmented into tissue types. After discarding the first five volumes, the remaining rs-fMRI images were motion and slice-time corrected, coregistered to T1, spatially normalized using the diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL) algorithm (Ashburner, 2007), linearly detrended, smoothed with an isotropic Gaussian filter with 8 mm3 full-width at half-maximum, corrected for 24 head movement parameters and physiological noise simulated by white matter and cerebrospinal fluid signals (Friston et al., 1996; Dagli et al., 1999; Windischberger et al., 2002), and temporally bandpass filtered (0.01–0.1 Hz).
The DTI images were preprocessed using FSL6.0 (fsl.fmrib.ox.ac.uk/fsl/), including motion and eddy current distortion corrections using a 12-degrees of freedom affine registration to the first B0 (b = 1,000 s/mm2) volume (Smith, 2002). Diffusion tensor models were estimated at each voxel, and diagonalization was performed to yield three eigenvalues and eigenvectors (Mori et al., 1999). T1 images were linearly coregistered to the B0 image in the DTI space and were further nonlinearly warped to the ICBM152 structural template in the Montreal Neurological Institute (MNI) space. The resulting inverse transformation matrix was used to warp the automated anatomical labeling (AAL) mask (Tzourio-Mazoyer et al., 2002) from the MNI space to the individual DTI native space (Gong et al., 2009).
Network construction
Brain networks were constructed with nodes representing brain regions and weighted edges depicting inter-regional connectivity. Nodes were defined by parcellating the cortex into 78 AAL atlas-based regions of interest (ROIs; 39 ROI pairs across hemispheres; Extended Data Table S1).
For the functional network, the mean time series of the 78 ROIs were extracted and intercorrelated, resulting in a 78 × 78 correlation matrix for each participant. The weighted edge of a node pair was defined as significant after Bonferroni-corrected p < 0.05.
For the structural network, parcellation of DTI images was performed in individual native spaces based on inversely warped AAL maps (see above, MRI data preprocessing), resulting in 78 ROIs for each participant. Voxel-based fiber tracking of the entire cortex was performed using PANDA (Cui et al., 2013; nitrc.org/projects/panda), with a fractional anisotropy (FA) threshold of 0.2 and a tracking turning angular threshold of 45° between two connections. Fiber connections were defined if at least three fiber bundles passed through or terminated in cortical ROIs (Li et al., 2009; Shu et al., 2009). The weighted edge of a node pair was obtained by multiplying the number of fiber bundles and the mean FA along the fiber bundles connecting a node pair, resulting in a 78 × 78 matrix for each participant (Gong et al., 2009; Lo et al., 2010).
The reproducibility of the network construction was examined using a range of network-formation thresholds, and consistent results were found for group comparisons of the topological properties (Extended Data Table S2).
Topological properties
Network-level properties
The network-level measures included (1) small-world parameters involving clustering coefficient
For a network G with N nodes,
The network's
Node-level properties
Nodal betweenness and nodal efficiency characterize the effectiveness of communication of a given node within a network. The betweenness of a given node
Nodal efficiency
PET image analyses
The PET images were preprocessed using PETPVE12 (Gonzalez-Escamilla et al., 2017; github.com/GGonEsc/petpve12) in SPM12. Images were realigned to T1 images, spatially normalized to the MNI space using the DARTEL algorithm (Ashburner, 2007), and partial volume effects corrected using the Müller–Gärtner method (Müller-Gärtner et al., 1992).
PET signal intensity was normalized to the average signal in the cerebellar gray matter using a non-PVE–corrected PET dataset (Gottesman et al., 2017). The cerebellum was masked using a 50% gray matter probability threshold for reference region selection. The global amyloid was calculated by averaging the standardized uptake value ratio (SUVr) of the entire cortex. The potential signal differences between the two PET scanners have been addressed by using the permutation test (Extended Data Text S1). A cutoff of mean SUVr > 1.2 was used to identify amyloid-positive cases (Gottesman et al., 2017). The regional amyloid burden was defined using 78 anatomical regions from both hemispheres based on the AAL atlas (see Extended Data Table S1). Regional SUVr values were computed by normalizing each region's mean amyloid uptake to the mean uptake of the reference region. To establish amyloid positivity at the regional level, we applied the same validated threshold of SUVR ≥ 1.2. Regions were classified as positive or negative based on this cutoff.
In line with previous autopsy (Braak and Braak, 1991) and PET studies (Grothe et al., 2017; Sakr et al., 2019), a frequency-based staging was applied, in which the frequency of amyloid positivity in a ROI is used as an indicator of the temporal involvement of that region in the course of spatial progression. Using the frequency-based method, we ranked 39 ROI pairs (averaged from two hemispheres) based on the proportion of participants showing amyloid positivity in each region. The amyloid burden was staged as follows: Stage 1, regions with >50% of participants classified as amyloid-positive; Stage 2, regions with 30–50% amyloid positivity; Stage 3, regions with 10–30% amyloid positivity; and Stage 4, regions with <10% amyloid positivity. This method provides a spatially progressive framework for amyloid deposition, reflecting its spread across brain networks rather than relying solely on global SUVR values. Finally, mean SUVr values were extracted from each anatomical division to form regional SUVr.
Statistical analysis
Sample characteristics
Independent two-sample t tests and χ2 tests were employed to assess sample characteristics (age, sex, years of education, ApoE genotype distribution, cognition, SCD features, HAMD, GDS, and HAMA scores) of the two groups. A p < 0.05 was considered significant.
Group comparisons of the topological properties
Both network- and node-level topological properties of the functional and structural brain network were defined as outcome measures of the current study. Generalized linear models (GLM) implemented in the “fitglm” function of MATLAB were used to assess group differences in topological properties accounting for age, sex, years of education, and ApoE genotype. For seven network-level properties, a Bonferroni-corrected p < 0.05 (0.05/7) was set as the statistic threshold. A false discovery rate (FDR)-corrected p < 0.05 was considered significant for node-level properties.
Associations between topological properties and behavioral scores
Pearson's correlations were carried out between topological properties and behavioral scores, including four episodic memory subscores, SCD-domain, and SCD-plus. A Bonferroni-corrected p < 0.05 (0.05/6) was considered significant.
Associations between topological properties and amyloid burden
In the SCD PET subgroup, Pearson's correlations between all topological properties and SUVr values (i.e., global SUVr and regional SUVr of four anatomical divisions) were calculated. For network-level properties, a statistic threshold of Bonferroni-corrected p < 0.05 (0.05/7) was applied. For node-level properties, a statistic threshold of FDR-corrected p < 0.05 was employed.
Results
Sample characteristics
The SCD group showed higher years of education, SCD-domain, SCD-plus scores, HAMD, GDS, and HAMA scores than the NC group. No other sample characteristic differences were found (Table 1). Table 1 also shows sample details of the SCD PET subgroup.
Demographic and neuropsychological assessments
Group comparisons of topological properties
We found group differences in node-level properties (p < 0.05, FDR-corrected; Extended Data Table S2) but not in network-level properties (p > 0.05, Bonferroni-corrected; Extended Data Table S2). For the functional network, SCD showed increased functional nodal efficiency (fNE) in the left olfactory cortex (OLF), the right insula (INS), bilateral anterior cingulate gyri (ACG), and left median cingulate gyrus (DCG). For the structural network, SCD showed increased structural nodal betweenness (sNB) in the left ACG and left DCG, as well as decreased sNB in the right superior medial frontal gyrus (Table 2; Fig. 1A). Due to the significant group differences in the psychiatric symptoms (Table 1), we additionally controlled for the HAMD, GDS, and HAMA scores and found that the group differences in network properties remained significant.
Topological differences and correlations in SCD. A, Altered node-level topological properties in SCD versus NC; left, increased fNE; right, altered sNB; regions showing significant effects were mapped onto the brain surface with the BrainNet Viewer (Xia et al., 2015); see Extended Data Table S1 for abbreviations. B, Bar plots for node-level topological property changes in the left anterior cingulate gyrus (ACG.L) and the left median cingulate gyrus (DCG.L); C, Positive correlations between SCD characteristics and the sNB in ACG.L. *p < 0.05, FDR-corrected.
Altered regional topological properties in the SCD group
Associations between topological properties and behavioral scores
Both SCD-domain and SCD-plus scores were positively associated with sNB in the left ACG in the total sample (p < 0.05, Bonferroni-corrected; Fig. 1C). No other associations were found (Extended Data Table S3).
Frequency-based amyloid staging in SCD
Among the subgroup of SCD (n = 55) with amyloid PET, four participants (7.3%) exceeded the global amyloid positivity cutoff (global SUVr > 1.2). The 39 ROI pairs were sorted by ROI-level amyloid positivity frequency and were stratified into four anatomical divisions according to frequency ranges (Fig. 2; Extended Data Table S4). Amyloid deposition was most frequently observed in the basal and lateral portion of the brain, including the FFG, inferior temporal gyrus (ITG), and middle temporal gyri (MTG; >50%; Anatomical Division 1), followed by regions within the occipital cortex, including the lingual, inferior, and middle occipital gyri (30–50%, Anatomical Division 2). The third division included the regions typically recognized as default mode network (DMN; Buckner et al., 2008), including the cingulate, supramarginal, angular, precuneal, parahippocampal gyri (PHG), as well as other midline brain areas (rectus and calcarine), and lateral frontotemporal regions (rolandic operculum and INS; 10–30%). The remaining areas (20 out of 39 ROI pairs) showed considerably less frequent amyloid depositions in SCD (<10%). The four-stage anatomical divisions yielded a highly consistent hierarchical deposition pattern across participants. Thirteen of the 55 SCD participants were negative on all stages. For the remaining participants with regional amyloid deposition, 100% (42/42) can be classified into any of the four stages (Extended Data Fig. S1). Fourteen subjects were positive at Stage 1 only; 10 subjects were positive at Stages 2 and 1; four subjects were positive at Stages 3, 2, and 1; four subjects were positive at Stages 4, 3, 2, and 1. In summary, individuals at more advanced stages (e.g., Stage 3 and Stage 4) are also positive in all earlier stages (e.g., Stage 1 and Stage 2).
Staging of amyloid deposition and network hub involvement in SCD. A, Results of the frequency-based amyloid staging. Frequency of ROI-level amyloid positivity across SCD participants on a color scale from magenta (highest) to yellow (lowest). The 39 ROI pairs are merged into four larger anatomical divisions. B, Bar plots showed the ROI-level amyloid positivity frequency. The color scale showed the frequency ranges: yellow (division 4, <10%), green (division 3, 10–30%), blue (division 2, 30–50%), and magenta (division 1, >50%). The circles and the rhombuses above the frequency bars indicate the hub regions within the structural or functional network, respectively. The solid circles and rhombuses represent the bilateral hubs, while the half-solid circles and rhombuses represent the unilateral hubs. See Extended Data Table S1 for abbreviations.
Association between topological properties and amyloid burden
Significant positive associations between the SUVr values and node-level topological properties were detected (p < 0.05, FDR-corrected; Fig. 3; Extended Data Fig. S2; Table 3).
Correlational analysis of amyloid burden and nodal topology in SCD. Pearson's correlations between global and regional amyloid load (global SUVr and regional SUVr values of Anatomical Divisions 1–4) and node-level topological properties. A, fNE; B, C, sNB. *p < 0.05, FDR-corrected. L, left; R, right; PCG, posterior cingulate gyrus; ACG, anterior cingulate gyrus; PoCG, postcentral gyrus.
Association between topological organizations and amyloid burden
For the functional network, the fNE of the left posterior cingulate gyrus (PCG) was consistently associated with the regional SUVr values of all amyloid cortical stages, as well as the global SUVr (Fig. 3A). The positive associations between the global SUVr and fNE of the right PCG and bilateral PHG were replicated by the associations between fNE of these nodes and the regional SUVr values from three out of four amyloid cortical stages (Extended Data Fig. S2A–C). In addition, the fNE of the right MTG was associated with regional amyloid retention of Stage 2 (Extended Data Fig. S2D).
For the structural network, the sNB of the left ACG and right postcentral gyrus were consistently associated with the regional SUVr values of all stages, as well as the global amyloid burden (Fig. 3B,C). The sNB of the right ACG was related to regional amyloid retention of Stage 2 (Extended Data Fig. S2F), and the sNB of the left superior parietal gyrus was related to regional amyloid retention of Stages 1 and 2 (Extended Data Fig. S2G).
No association between the network-level properties and the amyloid load was found (p > 0.05, Bonferroni-corrected).
Hub areas of the structural and functional networks
The hubs of the structural and functional networks were identified by the nodal betweenness using a cutoff of >1.5 times the mean nodal betweenness of the corresponding network (He et al., 2008). The structural hubs are distributed bilaterally, including the FFG, ITG, MTG, lingual gyrus, rectus, ACG, superior temporal gyrus, calcarine, rolandic operculum, and precuneus, despite two unilateral hubs (i.e., left DCG, right INS). All structural hubs belong to amyloid cortical Stages 1–3, not Stage 4 (Fig. 2B). On the contrary, the functional hubs are distributed mainly unilaterally and randomly across all four stages (Fig. 2B). A few areas were identified as both structural and functional hubs, including the bilateral FFG, the left superior temporal gyrus, the right ITG, the right INS, and the right precuneus.
Group comparisons showed increased nodal-level properties in SCD in the areas mainly overlapping with structural hubs, including bilateral ACG (structural hub), left DCG (structural hub), and right INS (structural and functional hub; Table 2). The regions showing consistent increases in both fNE and sNB are structural hubs (left ACG and left DCG). In the SCD PET subgroup, the amyloid burden was associated with the node-level properties of a set of structural or functional hubs (Table 3). The sNB of the structural hub left ACG and the fNE of the functional hub left PCG were associated with the global and regional SUVr across all cortical stages.
Discussion
The current study is the first to simultaneously assess structural and functional network topology in SCD. We found altered node-level (not network-level) topological changes in SCD compared with NC. Global and regional SUVr values across cortical amyloid stages (derived from frequency-based analyses) were positively associated with node-level topological changes in DMN hubs, such as the left ACG and bilateral PCG and PHG, suggesting that amyloid progression continuously influences network adaptation via increased information-processing burden in DMN hubs in SCD.
We observed increased node-level properties in SCD versus NC, predominantly in structural hubs (Table 2). The DMN nodes left cingulate (ACG, DCG; Buckner et al., 2008) showed consistent increases in both fNE and sNB, suggesting that increased structural betweenness may provide the structural basis for increased information exchange efficiency of the functional network. In agreement with our previous finding (Jiang et al., 2023), the sNB of the left ACG was correlated with SCD characteristics, reflecting increased self-awareness of cognition. Note that alterations of other node-level properties (i.e., SFGmed, INS, and OLF) occur either in the structural or functional network, suggesting that functional and structural reorganization may occur independently from each other. The reduced sNB in the SFGmed is consistent with a prior finding of reduced edge weights of this node in SCD (Kim et al., 2019). The increased fNE in the INS is in accordance with our previous finding of increased INS–hippocampal functional connectivity in SCD (Jiang et al., 2023). The increased fNE in the left OLF contrasts an earlier finding of decreased structural nodal efficiency in this area (Zhang et al., 2023), which warrants further study due to the small sample size of the latter study (nSCD = 13). The unchanged network-level topology in SCD is consistent with previous studies that have matched cognitive performance between groups and controlled for confounding factors such as age, sex, and education. Overall, the current study demonstrated increased information exchange efficiency in the DMN structural hub areas with preserved network-level efficiency in SCD.
Within the SCD PET subgroup, the ROI-level amyloid positivity rates ranged from 3.6 to 83.6% across 39 ROI pairs. Four of the 55 SCD participants (7.3%) were identified as amyloid-positive using global SUVr. Our previous analysis of this dataset has discussed several potential factors influencing the global positivity rate, such as age, source of recruitment, and ethnicity (Jiang et al., 2023). Consistent with previous studies (Grothe et al., 2017; Jelistratova et al., 2020), a substantial proportion of the participants (29/42, 69%) showed amyloid deposition in early anatomical divisions (Stages 1–3) despite being classified as amyloid-negative by global SUVr. Thus, staging analysis provides a more refined characterization of the amyloid burden than the global measure (Fantoni et al., 2020).
The amyloid staging results replicated several previous reports of the earliest retention in the basal part of the brain (Stage 1; Braak and Braak, 1991; Grothe et al., 2017; Sakr et al., 2019; Jelistratova et al., 2020; Levin et al., 2021). However, other studies have reported initial deposition in midline cortical structures (Villeneuve et al., 2015; Palmqvist et al., 2017; Collij et al., 2020) or lateral temporoparietal region (Villain et al., 2012; Guo et al., 2020). Various staging methods and clinical populations studied may contribute to different staging patterns. We found a higher positivity rate in the occipital areas (Stage 2) compared with the midline area ACG, which is consistent with a previous staging result in SCD (Sakr et al., 2019). A reversed pattern (ACG prior to occipital lobe) has been observed in studies across the AD spectrum (Grothe et al., 2017; Collij et al., 2020). Interestingly, a recent large-scale study (>3,000 cases) of the AD continuum identified different subtypes of amyloid progression trajectories, with earlier deposition in either the occipital lobe or anterior/posterior midline areas (Collij et al., 2022). It is unknown whether our result (occipital lobe prior to ACG) represents a typical progression trajectory in SCD (Sakr et al., 2019) or represents a subtype of AD pathological trajectory (Collij et al., 2022). Importantly, SCD is a broad condition and may present different types of clinical progression (reversible SCD, stable SCD, or SCD with subsequent cognitive decline; Jessen et al., 2020). Likewise, individuals with early-stage amyloid deposition may subsequently progress to a higher stage, remain stable, or revert to a lower stage (Jelistratova et al., 2020). Future longitudinal studies are needed to investigate the congruence between the clinical conversion pattern and the amyloid progression pattern in SCD.
When spatially comparing the network hubs with the staging results (Fig. 2B), all structural hubs belong to early-stage regions (Stages 1–3), the majority of which are bilaterally distributed. In contrast, functional hubs are unilaterally distributed and randomly located across stages. Our findings partly support previous studies showing spatial overlaps between critical amyloid deposition sites and the functional DMN (Buckner et al., 2009; Palmqvist et al., 2017), while we observed more spatial overlap between early amyloid stages and structural hubs than functional hubs, suggesting the white matter structural network may play a crucial role in early amyloid deposition (Depp et al., 2023).
We found higher amyloid burden is associated with increased nodal efficiency in structural hubs (bilateral ACG, right MTG) and functional hubs (bilateral PCG, bilateral PHG), all of which are pivotal nodes within the DMN (Buckner et al., 2008), which contrasts our previous findings using the same dataset showing negative associations between global amyloid and functional connectivity of DMN nodes (Jiang et al., 2023). Indeed, previous studies found both hyper- and hypoconnectivity in association with amyloid burden within the same cohort (Mormino et al., 2011; Elman et al., 2014; Lim et al., 2014; Jones et al., 2016). Unlike earlier connectivity-focused studies, the current graph-theoretic analysis characterizes system-level network alterations and reveals the efficiency or relative importance of a given node with respect to the whole brain network. Our results suggest early amyloid deposition in SCD is related to increasing nodal efficiency of the DMN nodes, indicating an increasing information-processing burden in relation to higher amyloid burden (Jones et al., 2016). It may also reflect a survival bias of high-amyloid individuals, such that individuals with higher nodal properties may serve as a protective factor, being more resilient to excessive processing demands of the DMN hubs (Schultz et al., 2017). The latter argument goes particularly well with the positive sNB–amyloid association in the structural hub ACG.L (Table 3), which may serve as the structural basis for the maintenance of functional efficiency and preservation of normal cognition in SCD. On the other side, increasing functional efficiency at the functional hubs (bilateral PCG, PHG.R) in association with amyloid load suggests that functional reorganization may occur independently of structural hubs. These findings indicate the complexity of the structural–functional relationship.
Our results suggest the importance of nodal property alterations in the core DMN structure cingulate cortex, as it shows increases in SCD compared with NC, as well as in association with amyloid pathology in SCD. The consistent topology–amyloid associations across all cortical stages suggest that the complex brain network reorganization via the core DMN structures already occurs during the initial amyloid retention at remote brain areas (as early as Stage 1 retention at the basolateral part of the brain). This phenomenon is consistent with the previous notion that system-level network alterations may occur prior to the characteristic spread of molecular pathology (Jones et al., 2016). Our data further suggest that the system-level network adaptation does not require the spatial proximity of amyloid pathology in the brain (Jones et al., 2016). It seems that amyloid pathology is associated with large-scale network reorganization via DMN hub structures rather than reflecting local alterations of the immediate site of the amyloid deposition (Jones et al., 2016, 2017). The increased nodal properties in these DMN hubs may reflect the increased information-processing burden along with the elevated amyloid deposition, contributing to the maintenance of normal cognition in SCD. Alternatively, hyperactivated DMN may actively stimulate the spreading of pathological proteins (Vogel et al., 2023; Giorgio et al., 2024).
This study highlights network topology changes in SCD and their associations with amyloid burden. While network-level properties remained unchanged, altered node-level properties and their links to amyloid suggest early AD-related neural adaptations in this at-risk population. Increased information exchange efficiency in DMN hubs with amyloid progression points to a growing information-processing burden. However, the study has limitations: only baseline data from the SILCODE study were used for amyloid staging, which may introduce instability, and the PET subgroup had a small sample size, preventing APOE-specific subgroup or interaction analyses. The lack of an NC subgroup with amyloid PET prevented us from directly comparing amyloid-related network changes between SCD and NC. Additionally, findings are based on a Chinese SCD cohort, limiting generalizability due to potential cultural and ethnic differences in brain connectivity and amyloid associations (Jiang et al., 2023). Future longitudinal and cross-ethnic comparison studies are needed to validate the utility of frequency-based staging in the early detection of amyloid deposition, the progression trajectory, its relationship to neural network adaptation, and the predictive values for the clinical progression of SCD.
Footnotes
The authors thank the individuals participating in the study and their families. Data used in the preparation of this article were obtained from the Sino Longitudinal Study on Cognitive Decline (SILCODE) Study. The authors thank coinvestigators who contributed to the design and implementation of the study. This work was supported by STI2030-Major Projects 2021ZD0200900 and 2022ZD0211800; Hainan Key Research and Development Project ZDYF2021SHFZ049; National Natural Science Foundation 32300864, 82020108013, and 82327809; Koeln Fortune Program; Sino-German Cooperation Grant M-0759; Shenzhen Bay Scholars Program Initiative Funding of Hainan University (KYQD-ZR-21057); and Tianchi Scholars Program.
↵*X.H. and Y.H. contributed equally to this work.
The authors declare no competing financial interests
This paper contains supplemental material available at: https://doi.org/10.1523/JNEUROSCI.2310-24.2025
- Correspondence should be addressed to Xiaochen Hu at xiaochen.hu{at}uk-koeln.de or Feng Yue at fyuee{at}hotmail.com.
